tcp/dccp: fix another race at listener dismantle
authorEric Dumazet <edumazet@google.com>
Thu, 18 Feb 2016 13:39:18 +0000 (05:39 -0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 18 Feb 2016 16:35:51 +0000 (11:35 -0500)
Ilya reported following lockdep splat:

kernel: =========================
kernel: [ BUG: held lock freed! ]
kernel: 4.5.0-rc1-ceph-00026-g5e0a311 #1 Not tainted
kernel: -------------------------
kernel: swapper/5/0 is freeing memory
ffff880035c9d200-ffff880035c9dbff, with a lock still held there!
kernel: (&(&queue->rskq_lock)->rlock){+.-...}, at:
[<ffffffff816f6a88>] inet_csk_reqsk_queue_add+0x28/0xa0
kernel: 4 locks held by swapper/5/0:
kernel: #0:  (rcu_read_lock){......}, at: [<ffffffff8169ef6b>]
netif_receive_skb_internal+0x4b/0x1f0
kernel: #1:  (rcu_read_lock){......}, at: [<ffffffff816e977f>]
ip_local_deliver_finish+0x3f/0x380
kernel: #2:  (slock-AF_INET){+.-...}, at: [<ffffffff81685ffb>]
sk_clone_lock+0x19b/0x440
kernel: #3:  (&(&queue->rskq_lock)->rlock){+.-...}, at:
[<ffffffff816f6a88>] inet_csk_reqsk_queue_add+0x28/0xa0

To properly fix this issue, inet_csk_reqsk_queue_add() needs
to return to its callers if the child as been queued
into accept queue.

We also need to make sure listener is still there before
calling sk->sk_data_ready(), by holding a reference on it,
since the reference carried by the child can disappear as
soon as the child is put on accept queue.

Reported-by: Ilya Dryomov <idryomov@gmail.com>
Fixes: ebb516af60e1 ("tcp/dccp: fix race at listener dismantle phase")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/inet_connection_sock.h
net/dccp/ipv4.c
net/dccp/ipv6.c
net/ipv4/inet_connection_sock.c
net/ipv4/tcp_ipv4.c
net/ipv6/tcp_ipv6.c

index 481fe1c9044cfd8b49585139e24df16b0716debf..49dcad4fe99e0ad5de491ef0e0675a3b516aabca 100644 (file)
@@ -270,8 +270,9 @@ struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
                                            struct sock *newsk,
                                            const struct request_sock *req);
 
-void inet_csk_reqsk_queue_add(struct sock *sk, struct request_sock *req,
-                             struct sock *child);
+struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
+                                     struct request_sock *req,
+                                     struct sock *child);
 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
                                   unsigned long timeout);
 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
index 5684e14932bd47e97b9d547307bfc50230e7be7d..902d606324a04ff81adbdc301a1ed58baa3f61f9 100644 (file)
@@ -824,26 +824,26 @@ lookup:
 
        if (sk->sk_state == DCCP_NEW_SYN_RECV) {
                struct request_sock *req = inet_reqsk(sk);
-               struct sock *nsk = NULL;
+               struct sock *nsk;
 
                sk = req->rsk_listener;
-               if (likely(sk->sk_state == DCCP_LISTEN)) {
-                       nsk = dccp_check_req(sk, skb, req);
-               } else {
+               if (unlikely(sk->sk_state != DCCP_LISTEN)) {
                        inet_csk_reqsk_queue_drop_and_put(sk, req);
                        goto lookup;
                }
+               sock_hold(sk);
+               nsk = dccp_check_req(sk, skb, req);
                if (!nsk) {
                        reqsk_put(req);
-                       goto discard_it;
+                       goto discard_and_relse;
                }
                if (nsk == sk) {
-                       sock_hold(sk);
                        reqsk_put(req);
                } else if (dccp_child_process(sk, nsk, skb)) {
                        dccp_v4_ctl_send_reset(sk, skb);
-                       goto discard_it;
+                       goto discard_and_relse;
                } else {
+                       sock_put(sk);
                        return 0;
                }
        }
index 9c6d0508e63a2ab7f13105cd91ea8c027c4a7557..b8608b71a66d34894722c17121754c403c74d946 100644 (file)
@@ -691,26 +691,26 @@ lookup:
 
        if (sk->sk_state == DCCP_NEW_SYN_RECV) {
                struct request_sock *req = inet_reqsk(sk);
-               struct sock *nsk = NULL;
+               struct sock *nsk;
 
                sk = req->rsk_listener;
-               if (likely(sk->sk_state == DCCP_LISTEN)) {
-                       nsk = dccp_check_req(sk, skb, req);
-               } else {
+               if (unlikely(sk->sk_state != DCCP_LISTEN)) {
                        inet_csk_reqsk_queue_drop_and_put(sk, req);
                        goto lookup;
                }
+               sock_hold(sk);
+               nsk = dccp_check_req(sk, skb, req);
                if (!nsk) {
                        reqsk_put(req);
-                       goto discard_it;
+                       goto discard_and_relse;
                }
                if (nsk == sk) {
-                       sock_hold(sk);
                        reqsk_put(req);
                } else if (dccp_child_process(sk, nsk, skb)) {
                        dccp_v6_ctl_send_reset(sk, skb);
-                       goto discard_it;
+                       goto discard_and_relse;
                } else {
+                       sock_put(sk);
                        return 0;
                }
        }
index 46b9c887bede0568ec378d3b809dda8fa463a166..64148914803a8443ecc0de2a45c141ae72cc0258 100644 (file)
@@ -789,14 +789,16 @@ static void inet_child_forget(struct sock *sk, struct request_sock *req,
        reqsk_put(req);
 }
 
-void inet_csk_reqsk_queue_add(struct sock *sk, struct request_sock *req,
-                             struct sock *child)
+struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
+                                     struct request_sock *req,
+                                     struct sock *child)
 {
        struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
 
        spin_lock(&queue->rskq_lock);
        if (unlikely(sk->sk_state != TCP_LISTEN)) {
                inet_child_forget(sk, req, child);
+               child = NULL;
        } else {
                req->sk = child;
                req->dl_next = NULL;
@@ -808,6 +810,7 @@ void inet_csk_reqsk_queue_add(struct sock *sk, struct request_sock *req,
                sk_acceptq_added(sk);
        }
        spin_unlock(&queue->rskq_lock);
+       return child;
 }
 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
 
@@ -817,11 +820,8 @@ struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
        if (own_req) {
                inet_csk_reqsk_queue_drop(sk, req);
                reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
-               inet_csk_reqsk_queue_add(sk, req, child);
-               /* Warning: caller must not call reqsk_put(req);
-                * child stole last reference on it.
-                */
-               return child;
+               if (inet_csk_reqsk_queue_add(sk, req, child))
+                       return child;
        }
        /* Too bad, another child took ownership of the request, undo. */
        bh_unlock_sock(child);
index c84477949d3ac844e7cc0e7d8bfd5ca9df91b951..487ac67059e237bab05d2291dd8e0a82f26e2bb0 100644 (file)
@@ -1597,30 +1597,30 @@ process:
 
        if (sk->sk_state == TCP_NEW_SYN_RECV) {
                struct request_sock *req = inet_reqsk(sk);
-               struct sock *nsk = NULL;
+               struct sock *nsk;
 
                sk = req->rsk_listener;
                if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
                        reqsk_put(req);
                        goto discard_it;
                }
-               if (likely(sk->sk_state == TCP_LISTEN)) {
-                       nsk = tcp_check_req(sk, skb, req, false);
-               } else {
+               if (unlikely(sk->sk_state != TCP_LISTEN)) {
                        inet_csk_reqsk_queue_drop_and_put(sk, req);
                        goto lookup;
                }
+               sock_hold(sk);
+               nsk = tcp_check_req(sk, skb, req, false);
                if (!nsk) {
                        reqsk_put(req);
-                       goto discard_it;
+                       goto discard_and_relse;
                }
                if (nsk == sk) {
-                       sock_hold(sk);
                        reqsk_put(req);
                } else if (tcp_child_process(sk, nsk, skb)) {
                        tcp_v4_send_reset(nsk, skb);
-                       goto discard_it;
+                       goto discard_and_relse;
                } else {
+                       sock_put(sk);
                        return 0;
                }
        }
index 1a5a70fb85512bb607cc8faefc6799e5340a4064..5c8c842730286ea4212222769d124f9e3d611333 100644 (file)
@@ -1387,7 +1387,7 @@ process:
 
        if (sk->sk_state == TCP_NEW_SYN_RECV) {
                struct request_sock *req = inet_reqsk(sk);
-               struct sock *nsk = NULL;
+               struct sock *nsk;
 
                sk = req->rsk_listener;
                tcp_v6_fill_cb(skb, hdr, th);
@@ -1395,24 +1395,24 @@ process:
                        reqsk_put(req);
                        goto discard_it;
                }
-               if (likely(sk->sk_state == TCP_LISTEN)) {
-                       nsk = tcp_check_req(sk, skb, req, false);
-               } else {
+               if (unlikely(sk->sk_state != TCP_LISTEN)) {
                        inet_csk_reqsk_queue_drop_and_put(sk, req);
                        goto lookup;
                }
+               sock_hold(sk);
+               nsk = tcp_check_req(sk, skb, req, false);
                if (!nsk) {
                        reqsk_put(req);
-                       goto discard_it;
+                       goto discard_and_relse;
                }
                if (nsk == sk) {
-                       sock_hold(sk);
                        reqsk_put(req);
                        tcp_v6_restore_cb(skb);
                } else if (tcp_child_process(sk, nsk, skb)) {
                        tcp_v6_send_reset(nsk, skb);
-                       goto discard_it;
+                       goto discard_and_relse;
                } else {
+                       sock_put(sk);
                        return 0;
                }
        }